221 results about "Mixed model" patented technology

The invention provides an interaction method and system based on deep learning, and relates to the technical field of computer vision, and the method comprises the steps: collecting a continuous image sequence, extracting feature points, calculating the position variation and gradient variation of the feature points, constructing a two-dimensional feature vector, and carrying out the clustering through a Gaussian mixture model, thereby obtaining a stable feature class. And weighting the feature points based on a feature weight distribution function, calculating feature similarity and constructing a matching relationship. And calculating a dynamic parallax threshold in combination with the vehicle movement speed, screening reliable matching feature pairs, and solving pose transformation parameters. A feature change sequence is extracted through a feature detection network, a motion feature mapping matrix is constructed, and a compensation depth map is generated in combination with depth information. And scene reconstruction is carried out based on the compensation depth map and the pose transformation parameters, the scene confidence score is calculated, the pose parameters are calibrated, a high-precision pose parameter sequence is obtained, and the robustness of feature matching and the accuracy of pose estimation can be improved.

The present invention discloses a coal mine disaster simulation prediction system based on multi-dimensional data collection, which relates to the field of coal mine safety technology. A variety of monitoring equipment such as geological radar and microseismic monitoring are integrated to build a real-time monitoring network covering key areas of coal mines, and comprehensively collect multi-dimensional data. The system includes modules such as data preprocessing, model construction and training, prediction and decision-making, early warning and emergency response, and user interaction interface. Data is processed through data cleaning, filtering, standardization and fusion technology, and disaster prediction is performed using a CNN and LSTM hybrid model to output disaster probability and risk areas, and automatically trigger early warning and emergency plans. The user interaction interface displays real-time monitoring, prediction results and early warning information, providing a comprehensive and intelligent solution for coal mine disaster prevention and control.

The invention discloses an automobile seat framework machining defect detection method based on machine vision, and particularly relates to the technical field of defect detection. The method comprises the following steps: constructing a multi-angle image acquisition and edge reflection modeling module aiming at complex defects such as weld joint pseudo soldering, microcracks, hole site deviation and collapse deformation, extracting weld joint continuity, edge integrity and hole site geometric consistency characteristics by using a deep neural network, and generating a structural feature vector; defect type recognition and credibility scoring are completed through small sample anomaly modeling and Gaussian mixture model classification, sub-pixel-level coordinate labeling of defect positions is achieved in combination with a Gaussian fitting algorithm, a defect positioning map is output, traceability analysis and severity grading are conducted based on historical data comparison, and the defect positioning accuracy is improved. The method is suitable for industrial online detection and quality closed-loop control.

The invention discloses an industrial equipment defect detection method based on multi-modal feature fusion and dynamic optimization, and belongs to the technical field of computer vision, and the method comprises the steps: 1, multi-modal industrial data collection: deploying multiple sensors in a production line, collecting data in multiple periods, constructing a defect-free and multi-type defect sample library, and carrying out multi-modal industrial data collection; a time-space aligned multi-modal label is marked; step 2, data enhancement and defect synthesis; step 3, multi-modal hybrid model training: constructing a hybrid network, and performing pre-training and fine tuning by using a dynamic loss function; step 4, edge end dynamic optimization and deployment: edge end reasoning is realized through dynamic knowledge distillation, and model fine tuning is automatically triggered when false detection and missing detection are found; and step 5, intelligent labeling and result visualization: a front-end interface displays a detection result in real time. The problem that a current target detection framework is not high in small target recognition accuracy and low in efficiency is solved, and the reliability of industrial equipment defect detection is improved.

The invention provides a magnetic target state estimation system and device, and relates to the technical field of magnetic target state estimation, and the system mainly comprises a data collection module, a magnetic field model construction module, a deep learning module and a result output module. The data acquisition module comprises a field measurement unit and a data set unit; the magnetic field model building module is used for enabling a magnetic target to be equivalent to a physical model; the physical model is a mixed model of an ellipsoid and a magnetic dipole array; the deep learning module comprises a neural network, the physical model and an optimizer; and the result output module is used for sending out the state prediction result of the magnetic target after the neural network training is completed. According to the scheme, the complex magnetic field data is efficiently fitted through the neural network, and the physical constraint is introduced into the loss function, so that the robustness is improved, and the overfitting problem in simple data fitting is avoided; the calculation time and resources can be greatly reduced; and the solution quality and stability can be improved.

The embodiment of the invention provides a multi-modal recognition model training method, a multi-modal data recognition method and related equipment. The multi-modal recognition model training method comprises the steps that multiple multi-modal training sample pairs are acquired; based on a multi-granularity output model in the multi-modal recognition model, performing first feature extraction and second feature extraction on each multi-modal training sample pair to obtain first feature data and second feature data with different corresponding processing granularities, and performing mixing processing on each first feature data and the corresponding second feature data to obtain a first feature data and a second feature data; multi-granularity mixed output is obtained; performing feature level mixing processing on each multi-modal training sample pair based on a feature level mixing model in the multi-modal recognition model to obtain feature level mixing output corresponding to each multi-modal training sample; the multi-modal loss is calculated based on the multi-granularity mixed output and the feature-level mixed output, and the multi-modal recognition model is updated based on the multi-modal loss, so that the recognition accuracy of the multi-modal data is effectively improved.

The invention discloses a GMM-L2 fault diagnosis method, device and equipment based on a DBGSA algorithm and a medium, and relates to the technical field of engine fault detection, and the method comprises the steps: obtaining the original data of a wind power engine; processing the original data of the wind power engine to obtain second data; then calculating a first clustering center in the second data for generating a first Gaussian mixture model for calculating a health degree; setting an alarm limit according to the health degree, and if the second data is smaller than the alarm limit, determining that the second data is in an abnormal state; processing the second data in the abnormal state to obtain third data; updating the clustering center of the third data by adopting a DBGSA algorithm of fuzzy clustering to obtain a third clustering center; and setting a membership threshold value of the fault type, and calculating the membership probability of the fault type to which each cluster belongs to obtain the fault type to which the third data belongs. According to the method, missing detection caused by classifying only one data point into one fault can be avoided.

Methods and systems implement a pleno-generation face video compression framework with bandwidth intelligence for generative models and compression. Heterogeneous-granularity facial description regularizes long-term dependencies between video frames and compensates for motion estimation errors caused by compact representations of motion information. A generative decoder reconstructs heterogeneous-granularity visual representations, providing auxiliary visual signals for attention-based recalibration of a GFVC-reconstructed face signal. A coarse-to-fine generation strategy avoids error accumulation. High efficiency for heterogeneous-granularity signal compression is achieved by two different entropy-based signal compression methods: heterogeneous-granularities feature representation from the key-reference frame as hyperpriors to optimize the entropy model for compressing heterogeneous-granularity feature from subsequent inter frames, and a feature difference operation for heterogeneous-granularities feature representation between key-reference and subsequent inter frames, such that the entropy model only compresses heterogeneous-granularities feature residual for redundancy reduction. Mixed-model dataset generation and training and model-specific dataset generation and training are also provided.

The invention provides a landslide disaster dynamic monitoring method based on a time sequence and a hybrid model, and belongs to the technical field of intelligent monitoring, and the method comprises the steps: collecting multi-source heterogeneous data of historical landslide, carrying out the preprocessing of the multi-source heterogeneous data, obtaining first data, carrying out the type division of the first data, and carrying out the time sequence coding of each type of data, obtaining single samples under the same time sequence to obtain a sample set, and training the neural network model according to the sample set to obtain an initial model; performing model verification on the initial model, and optimizing to obtain a time sequence and a hybrid model in combination with the constructed mapping relationship between the landslide deformation and the monitoring dimension under the corresponding type; and monitoring and acquiring current heterogeneous data of the target monitoring point, inputting the current heterogeneous data into the time sequence and the hybrid model, predicting to obtain a current disaster result of the target monitoring point, and outputting a prompt. The accuracy of the model is ensured through model verification and mapping relation construction, and a convenient basis is provided for landslide disaster prediction.

The invention discloses a text image pet re-identification method based on noise learning, and the method comprises the following steps: collecting pet image and description text pair data, and carrying out the standardization processing; vectorizing the image data set and the description text data set respectively; performing global and local similarity calculation on the image-text data set; training a cross-modal retrieval model; and performing model reasoning and retrieval. Through the Gaussian mixture model and the consensus division strategy, the loss function is subjected to smooth optimization, the robustness of noise data is enhanced, fine-grained feature learning and attention to a positive sample are supported, the noise data are effectively filtered, a label is calibrated, the robustness and matching precision of the noise data in a cross-modal task are improved, and the matching accuracy of the noise data is improved. Higher flexibility and theoretical guarantee are provided; meanwhile, through cross-modal feature extraction and noise robustness design, pet images under different illumination, backgrounds and postures can be processed, and the generalization ability in practical application is improved.

The invention discloses a straw coverage monitoring method based on multi-source remote sensing data, and the method comprises the steps: carrying out the interpolation processing of SMAP brightness temperature data and ASCAT backscattering data through a Backus-Gilbert method, and generating down-sampling data with the resolution of 10 km; the method comprises the following steps: up-sampling Sentinel-1 synthetic aperture radar data, MODIS NDVI data and LST data to a resolution ratio of 10km; according to the Sentinel-1 data, the MODIS NDVI data and the LST data after up-sampling, a regression model of the 10 km ASCAT data is constructed; linear analysis is carried out according to the vegetation water content and the total primary productivity, VWC-VOD and GPP-VOD fitting equations are obtained, and intercept values of the fitting equations are extracted; defining the intercept value as the inverse evolution of the straw coverage; and constructing a linear hybrid model based on the VOD feature vectors of the end members, and inverting the abundance of the end members through minimum error measurement to obtain the farmland straw coverage. According to the method, SMAP brightness temperature data with low resolution is improved to 1km resolution through a multi-step downscaling and inversion method, and the spatial detail expression capability of microwave remote sensing data is improved.

The present invention provides a method for detecting abnormal sound of multiple types of machines based on a diffusion model, comprising the following steps: collecting normal sound data, wherein the normal sound data contains category information, and constructing a training data set; preprocessing the normal sound data in the training data set to generate a mel-spectrogram, and generating a label corresponding to the category; training a multi-class abnormal sound detection network based on a diffusion model according to the label corresponding to the category and the mel-spectrogram; using the trained multi-class abnormal sound detection network to make mean‑GMM features, and training a Gaussian mixture model; inputting the sound data to be detected into the multi-class abnormal sound detection network and the Gaussian mixture model to obtain a detection result. The present invention designs an abnormal score algorithm based on a Gaussian mixture model, which evenly considers the continuity and change characteristics of the sound data in the time domain. A more fine-grained abnormal score is provided, which significantly improves the accuracy of detection and the adaptability to complex industrial environments.

The invention discloses an aero-engine gas path sensor fault diagnosis method based on GCN and BiLSTM, and belongs to the field of aero-engine health management. In order to solve the problems that in the prior art, sensor faults and gas path system fault signals are difficult to distinguish, and time sequence modeling is insufficient, a mixed model architecture fusing a graph convolutional network and a bidirectional long-short-term memory network is provided. The method comprises the following steps: screening high-correlation sensors through grey correlation degree to construct multiple groups of input-output combinations; the method comprises the following steps: extracting spatial topological characteristics among sensors by using a GCN, capturing bidirectional time sequence dependence by combining BiLSTM, and establishing a parallel prediction model cluster; and setting a dynamic threshold based on a reconstruction residual 3 sigma criterion, and realizing fault code matching through a Boolean diagnosis dictionary. The method effectively solves the problem of similarity between the sensor fault and the gas path system fault signal, can accurately diagnose the sensor fault or the gas path system fault, and improves the fault detection and maintenance efficiency of the aero-engine.

The invention discloses an interlayer gold purity evaluation method and device based on a hybrid model, equipment and a medium. The method comprises the following steps: collecting original data; inputting the processed original data into a hybrid model for training to obtain a trained hybrid model; wherein the hybrid model comprises a feature extraction module, a weighted fusion unit and a classification module which are connected in sequence; collecting a second pulsed eddy current signal and a second supplementary feature corresponding to the gold to be detected; inputting the second pulse eddy current signal and the second supplementary feature into a trained hybrid model, and outputting a prediction vector; and converting the prediction vector into a one-hot code, and obtaining the number of the corresponding gold to be tested through the one-hot code. The problem that an existing classification mixing model cannot be suitable for a highly-adulterated gold purity detection scene is solved, the classification precision of adulterated gold purity detection is improved, and the application range of the classification mixing model in the field of scarce sample detection is widened.

The invention relates to a meteorological-based dynamic graph network photovoltaic power station group ultra-short-term prediction method, which is characterized in that a dynamic space-time graph network is constructed, photovoltaic power stations in a region are regarded as a complex network system connected by meteorological fluctuations, each power station is taken as a node, and the propagation relationship of the meteorological fluctuations among the power stations is represented by edges of a graph. The weight and time delay of the edge are dynamically adjusted according to real-time meteorological data, and dynamic interaction between power stations, the overall trend under the stable meteorological condition and rapid fluctuation caused by sudden weather events are accurately captured in combination with a mixed model of a graph neural network and a recurrent neural network. Compared with a traditional method, the method has the advantages that the prediction precision, the real-time response capability and the physical interpretability are remarkably improved, scattered photovoltaic power stations are integrated into a mutually associated dynamic system, an efficient and reliable solution is provided for power prediction of a power station group, and the operation stability and the power grid dispatching efficiency of a large-scale photovoltaic system are enhanced.

The invention relates to a forged voice detection method adopting two-dimensional map attention, and the method comprises the steps: carrying out the pre-emphasis processing of an original voice, and extracting the linear frequency cepstrum coefficient features of each frame; gMM modeling is carried out on real voice and forged voice by using a BP algorithm according to linear frequency cepstrum coefficient features, iterative operation is carried out on training data by using an expectation maximization algorithm until convergence, the number of voice frames is counted, the LGP of each frame of voice feature in each Gaussian component is calculated, a constant term is subtracted, and standardization processing is carried out; according to the method, the Gaussian mixture model is modeled by adopting a bidirectional subdivision method (BP), so that the difference of mean vectors of adjacent Gaussian components is relatively small, the dependency relationship between the Gaussian components can be effectively captured by two-dimensional convolution, a clearer and tighter feature relationship is provided for a graph attention module, and the image attention is more accurate. According to the method, the feature space structure is optimized from local to global, and the performance of forged voice detection is improved.

The invention provides a mobile formwork intelligent monitoring method and system based on digital twinning and multi-modal perception, and belongs to the technical field of building construction intelligence, and the method comprises the steps: building a multi-modal sensor network on a mobile formwork entity; establishing a digital twin platform: constructing a BIM model mapped with a mobile formwork entity, pre-burying a virtual monitoring point in the BIM model, performing matching calibration with a sensor of the multi-modal sensor network, and fusing multi-source sensor data acquired by the multi-modal sensor network in real time; calculating a construction health index (SHI) in real time based on the multi-source sensor data; on the basis of the multi-source sensor data, the instability probability is predicted in real time through an AE-LSTM mixed model; meanwhile, predicting a weight factor of a construction health index SHI; and according to the SHI and the instability probability, jointly evaluating whether a mobile formwork adjustment instruction is generated or not and sending out an early warning. The safety state of the movable formwork is controlled in real time, and the problem that the instability trend cannot be monitored in real time is solved.

The invention discloses a sea ice risk prediction method and device, computer equipment, a storage medium and a product. The method comprises the following steps: acquiring historical environment data and historical sea ice monitoring data; constructing a linear regression relation model between the target environment factor and the sea ice data and a time sequence analysis model of the sea ice data according to the historical environment data and the historical sea ice monitoring data; fusing the linear regression relation model and the time sequence analysis model to obtain a hybrid model; and acquiring real-time environment data, and generating sea ice prediction data by using the hybrid model. The sea ice generation and change rule can be better described by combining the explanation capability of linear regression on influence factors and the dynamic grasp of a time sequence analysis model on historical data, the advantages of different models can be effectively integrated by introducing a hybrid model, the prediction precision of sea ice risks and the capture capability of trends are improved, and the prediction efficiency is improved. Therefore, more accurate sea ice risk prediction under a complex environment condition is realized.

The invention discloses a time-space intelligent area rainfall classification weighting mixing quantitative forecasting method based on Beidou water vapor. The method comprises the following steps: 1, acquiring atmospheric precipitable water and meteorological historical data inverted by Beidou; 2, constructing and training a BPNN rain-free model and a BPNN rain model; 3, training based on the univariate meteorological parameters in combination with CNN and LSTM network models; 4, on the basis of CNN and LSTM network model input, constructing a BPNN rain-free hybrid model and a BPNN rain hybrid model, and training; and 5, subsequent area rainfall classification weighting mixing quantitative forecasting is carried out. The method is simple in steps and reasonable in design, weighting is carried out on the basis of predicted values of the BPNN rain-free model and the BPNN rain model and predicted values of the BPNN rain-free mixed model and the BPNN rain mixed model, subsequent quantitative rainfall forecasting of the to-be-studied area is achieved, and short-term rainfall forecasting precision is improved.

The invention discloses a medium and low voltage line abnormity identification method based on a deep neural network, and the method achieves the comprehensive coverage of key nodes of a power distribution network through the four-stage deployment of station-line-transformer-household, completely captures the abnormal signals at the tail end of a branch line, provides sufficient data support for deep feature learning, solves a problem of a conventional monitoring blind area, and improves the recognition precision of the power distribution network. A convolutional neural network (CNN) and long short-term memory (LSTM) network mixed model automatically extracts time-frequency domain joint features, manual feature design is not needed, subtle changes of transient waveforms are effectively captured, the anomaly identification precision is improved, the limitation that a traditional method depends on artificial experience is broken through, potential fault hidden dangers can be found in advance, sufficient processing time is won for operation and maintenance personnel, and the operation and maintenance efficiency is improved. Through rapid positioning, the fault processing period is shortened, the power failure time is reduced, the power supply reliability is improved, fuzzy C-means clustering is combined with topology analysis, accurate positioning of abnormal sections is achieved, the manual inspection workload is reduced, the fault processing efficiency is improved, and the operation and maintenance cost is reduced.

The invention discloses a diet health index prediction method fusing M-One coding and deep learning. The method comprises the steps of collecting diet data of a target user; preprocessing the collected data, and performing multi-dimensional influence matrix M-One coding on discrete features to generate a two-dimensional influence matrix; determining parameters of a three-dimensional data format according to the characteristics of the input data; constructing a CNN-LSTM model, performing convolution operation on the data by using a one-dimensional convolutional neural network 1D-CNN, and extracting features of time series data; optimizing the size of a CNN convolutional layer, the number of neurons of the two LSTM layers and the learning rate through a sparrow algorithm; according to the method, M-One coding and a mixed model of 1D-CNN and LSTM are fused, deep learning optimization of multi-dimensional features is realized, and a scientific basis is provided for user health management.

The invention provides a random-interval hybrid uncertainty structure reliability analysis method and a related device, and belongs to the technical field of structure reliability analysis. The method mainly comprises two improvements: 1, determining a selection mode of a sampling center of an important sampling density function of a kth layer, constructing the important sampling density function according to the determined sampling center of the important sampling density function of the kth layer, and extracting a random sample of the kth layer according to the constructed important sampling density function; and 2, constructing a Kriging model according to the function relationship between the POD coefficient and the interval variable, constructing a mixed Kriging model on the basis of the Kriging model according to the function relationship between the output and the interval variable, and carrying out adaptive learning on the mixed Kriging model. According to the method, the problem that the reliability analysis efficiency of the random-interval hybrid uncertainty structure is not high is solved.

The invention belongs to the technical field of unsupervised deep learning methods, and particularly relates to an unsupervised deep learning method based on synthetic data generation, and the method comprises the steps: searching a clustering center and a standard deviation of original unlabeled data through a Gaussian mixture model; bayesian optimization selects an optimal superposition weight for a Gaussian mixture model by measuring the difference between isotropic Gaussian spot generation data based on a Gaussian mixture model clustering center and a standard deviation and original data distribution; the Gaussian mixture model determines the clustering number and the standard deviation under the optimal superposition weight, and the isotropic Gaussian spots generate a large amount of synthetic data with labels and balance based on the clustering number and the standard deviation; the synthetic data is used for training a deep neural network model; and the trained deep neural network model executes a downstream task based on the initial label-free data. The sample equalization of the generated synthetic data is realized, the training of the deep neural network is assisted, and the model precision in an unsupervised learning scene is improved.

The invention discloses a main beam formwork erection elevation adjustment method based on machine learning, and belongs to the technical field of elevation adjustment. The method comprises the steps that a space-time database is constructed by collecting data of the construction period of a cable-free section; processing the time sequence data by adopting an LSTM-Transform hybrid model, extracting long-period characteristics, and outputting an initial prediction value of the elevation; analyzing nonlinear variables such as a cable force change rate, a sunlight gradient and a material age by using an XG-Boost algorithm, and outputting a compensation factor; fusing and generating a joint prediction value; constructing a reinforcement learning agent, and outputting an adjustment instruction by taking a construction stage as a state space, taking an elevation adjustment amount as an action space, minimizing deviation between a predicted value and a measured value and taking construction stability as a reward function; the formwork erecting elevation is adjusted according to the driving hydraulic system, and the database is updated in real time. According to the invention, through multi-dimensional data fusion and intelligent optimization, the elevation adjustment precision and stability are improved.

The invention discloses a transport case real-time positioning method and device based on multi-sensor fusion, and belongs to the technical field of Internet of Things and intelligent logistics. The method comprises the following steps: acquiring positioning parameters and motion state parameters of a transport case in real time through sensors such as a GPS (Global Positioning System) and an IMU (Inertial Measurement Unit), and carrying out noise filtering, timestamp synchronization and coordinate system unified preprocessing operation on data; an extended Kalman filtering and particle filtering mixed model is adopted, and a weight coefficient is dynamically adjusted according to the GPS confidence coefficient and the IMU confidence coefficient, so that adaptive data fusion is realized; transmitting a positioning result through 4G / 5G and LoRa dual-mode communication, and judging reliability based on a residual error to trigger calibration or switch a positioning mode; according to the method, the problems of insufficient precision, poor robustness and limited real-time performance of a single sensor in a complex environment are solved, centimeter-level positioning precision in the complex environment is realized, the anti-interference performance of positioning is improved, the real-time performance and low power consumption of data transmission are ensured, and a fusion result is prevented from being polluted by abnormal data.

In an embodiment, a system for digital engagement is disclosed that is configured to operate an application for a patient with dementia. The application includes at least one applied virtual environment. The system is further configured to evaluate input of the patient based on a mixed model for observing digital engagement, and to engage the patient with the application using dialogue delivered by at least one virtual avatar, the virtual avatar being supported by artificial intelligence and the dialogue being determined based on the evaluated input. The system is further configured to log and evaluate progress data of the patient and to provide feedback and support based on the evaluated progress data.

The invention discloses an urban space-time evolution analysis method based on multi-source historical data. The method comprises the following steps: 1) acquiring and digitally processing historical city maps, point-of-interest data and realistic paintings in multiple periods; 2) performing syntactic analysis on the urban road network and the water system by adopting a mixed model of topology and angle distance, and calculating core indexes such as integration degree and angle selection degree; 3) based on an optimization algorithm, carrying out normalization processing on the spatial syntactic index and the POI data; 4) extracting dynamic city information from the historical drawing; 5) performing statistical test on dynamic data extracted by drawing and a spatial syntactic analysis result; and 6) analyzing the space-time aggregation mode and function evolution of the public service facility by adopting a location quotient formula and kernel density estimation.By constructing a closed-loop research framework of multi-source data fusion, algorithm optimization and quantitative verification, a highly reliable scientific method is provided for deeply revealing an internal driving mechanism of urban form evolution.

The invention relates to the technical field of educational resources, in particular to a digital textbook generation resource optimization scheduling method and system based on big data, and the method comprises the steps: 1, data collection and preprocessing; step 2, feature modeling and demand prediction; step 3, resource matching and optimization; step 4, resource scheduling and balancing; 5, teaching material generation and verification; by introducing quantum state coding normalization and time sequence data filling based on the generative adversarial network, multi-source data can be processed more accurately, the data quality is improved, in addition, the knowledge mastery degree is calculated by constructing the multi-dimensional feature vector containing the emotion feature and the learning style feature and adopting the knowledge association model based on quantum entanglement, and the knowledge mastery degree is improved. And a learning path is predicted based on a mixed model of the graph attention network and the time sequence convolutional network, so that the requirements of learners can be known more comprehensively and deeply, and the accuracy and pertinence of resource matching are greatly improved.

The invention discloses a Yangtze River water level prediction method of a hybrid model, belongs to the technical field of water level prediction, and aims to solve the problems of difficult nonlinear characteristic modeling, high calculation complexity, insufficient long sequence dependency relationship capture and the like in Yangtze River water level prediction. According to the method, data of six water level stations including Zhonghuanggui, Yichang, Baishaobrain, Laolinggui, two Chen kou and yao port in Yangtze River are taken as input, and missing values are filled by adopting a similarity-based method; the DSC module is used for learning the independent dependency relationship and spatial characteristics of each water level station; modeling a dynamic dependency relationship of the water level station along with time change by utilizing a GRU module; and finally outputting a Yangtze River water level prediction result. According to the method, the water level prediction precision is remarkably improved, and reliable support is provided for flood control and disaster reduction, water resource management and shipping scheduling; the method has the advantages of high prediction precision, high calculation efficiency, wide application range and the like, can effectively cope with challenges of extreme weather and climate changes on water level prediction, and makes contributions to guarantee of life and property safety of people in a drainage basin and economic and social development.

The invention discloses an automatic driving takeover performance evaluation method and system considering multiple influence factors, and the method comprises the steps: obtaining multi-modal data and influence factor data in the whole process of switching a working mode of an automatic driving system in different traffic takeover scenes through constructing a driving data collection assembly, and obtaining the performance of the automatic driving takeover performance through a subjective weighting method and an objective weighting method. Calculating the optimal weight of each related characteristic index in the multi-modal data, constructing an automatic driving takeover performance comprehensive evaluation model to obtain the comprehensive score, and constructing a generalized linear hybrid model based on the comprehensive score and the influence factor data to obtain the comprehensive evaluation result of the automatic driving takeover performance. Evaluating the influence importance of different influence factors in the influence factor data on the comprehensive score of the takeover performance; according to the method, the multi-dimensional takeover performance comprehensive evaluation model based on the DEA model is constructed by acquiring the multi-modal data and extracting the characteristic indexes of the multi-modal data, so that the problem of single-dimensional limitation of an existing evaluation system is solved, and the actual score of the automatic driving takeover performance is accurately reflected.